DirectML Operators

This section details the operators supported by DirectML, the Microsoft DirectX Low-Level Machine Learning API. These operators form the building blocks for creating and executing machine learning graphs.

Core Operators

DML_ACTIVATION_ELU_OPERATOR

Exponential Linear Unit (ELU) activation function.

DML_ACTIVATION_HARD_SIGMOID_OPERATOR

Hard Sigmoid activation function.

DML_ACTIVATION_LINEAR_OPERATOR

Linear activation function (identity).

DML_ACTIVATION_MISH_OPERATOR

Mish activation function.

DML_ACTIVATION_RELU_OPERATOR

Rectified Linear Unit (ReLU) activation function.

DML_ACTIVATION_SCALED_ELU_OPERATOR

Scaled Exponential Linear Unit (SELU) activation function.

DML_ACTIVATION_SIGMOID_OPERATOR

Sigmoid activation function.

DML_ACTIVATION_SOFTMAX_OPERATOR

Softmax activation function.

DML_ACTIVATION_TANH_OPERATOR

Hyperbolic Tangent (Tanh) activation function.

DML_BATCH_NORMALIZATION_OPERATOR

Batch Normalization operator.

DML_CONVOLUTION_OPERATOR

2D Convolution operator.

DML_GEMM_OPERATOR

General Matrix Multiply (GEMM) operator.

DML_MAX_POOLING_OPERATOR

2D Max Pooling operator.

DML_SUM_OPERATOR

Element-wise addition of two tensors.

DML_ELEMENT_WISE_ADD_OPERATOR

Element-wise addition of two tensors.

DML_ELEMENT_WISE_MULTIPLY_OPERATOR

Element-wise multiplication of two tensors.

Tensor Manipulation

DML_REDUCE_OPERATOR

Reduces a tensor along specified dimensions.

DML_SLICE_OPERATOR

Extracts a slice from a tensor.

DML_TILE_OPERATOR

Tiles a tensor along specified dimensions.

Other Operators

DML_QUANTIZE_OPERATOR

Quantizes a tensor to a lower-precision format.

DML_DEQUANTIZE_OPERATOR

Dequantizes a tensor from a lower-precision format.